Hello,
While playing with Oracle 10gR2 RAC on OCFS2, we decided to run some
iozone benchmarks on a single node OCFS2 v. ext3 on the same LUNs. ext3
outperforms OCFS2 by 12 - 27% in the tests.
While we have not tested Oracle benchmarks like Swingbench on the two
filesystems yet, we are worried OCFS2 is going to cause an immediate
performance hit.
When running iozone to the shared OCFS2 on two nodes, performances gets
even worse. This doesn't bode well for running RAC on multiple nodes
and actually increasing performance vs. a single node on ext3.
Any comments on how we can ensure performance of RAC will scale as
needed on OCFS2?
/Brian/
--
Brian Long | | |
IT Data Center Systems | .|||. .|||.
Cisco Linux Developer | ..:|||||||:...:|||||||:..
Phone: (919) 392-7363 | C i s c o S y s t e m s
On Fri, May 05, 2006 at 10:13:37AM -0400, Brian Long wrote:> While playing with Oracle 10gR2 RAC on OCFS2, we decided to run some > iozone benchmarks on a single node OCFS2 v. ext3 on the same LUNs. ext3 > outperforms OCFS2 by 12 - 27% in the tests.Were you running with the -I option to ensure O_DIRECT I/O on all filesystems? Actually, does strace(1) prove that out? Joel -- "You look in her eyes, the music begins to play. Hopeless romantics, here we go again." Joel Becker Principal Software Developer Oracle E-mail: joel.becker at oracle.com Phone: (650) 506-8127
Brian Long wrote:>Hello, > >While playing with Oracle 10gR2 RAC on OCFS2, we decided to run some >iozone benchmarks on a single node OCFS2 v. ext3 on the same LUNs. ext3 >outperforms OCFS2 by 12 - 27% in the tests. > >You need to go into a bit more detail there. What's the problem: random reads? random writes? sequential writes? As long as my latency times for log sync writes don't get out of hand, I can live with even a frightening degredation in datafile read/write performance. With a magnified SGA, I don't plan on doing enough random IO to worry about random IO performance. My datafile IO drops dramatically when I add nodes to my cluster.>While we have not tested Oracle benchmarks like Swingbench on the two >filesystems yet, we are worried OCFS2 is going to cause an immediate >performance hit. > >Why not just test the app? I can do wicked things to the cluster interconnect that bear no resemblance to what actually happens in our app.>When running iozone to the shared OCFS2 on two nodes, performances gets >even worse. This doesn't bode well for running RAC on multiple nodes >and actually increasing performance vs. a single node on ext3. > >Any comments on how we can ensure performance of RAC will scale as >needed on OCFS2? > >/Brian/ > >